tree: git://git.cmpxchg.org/linux-mmotm.git master
head: fa5441daae8ad99af4e198bcd4d57cffdd582961
commit: 05447370dd1cb96635f9502de590442a34903ff1 [173/329] psi: cgroup support
reproduce:
# apt-get install sparse
git checkout 05447370dd1cb96635f9502de590442a34903ff1
make ARCH=x86_64 allmodconfig
make C=1 CF=-D__CHECK_ENDIAN__
sparse warnings: (new ones prefixed by >>)
kernel/sched/psi.c:153:18: sparse: incorrect type in initializer (different address spaces) @@ expected struct psi_group_cpu *cpus @@ got struct psi_group_struct psi_group_cpu *cpus @@
kernel/sched/psi.c:153:18: expected struct psi_group_cpu *cpus
kernel/sched/psi.c:153:18: got struct psi_group_cpu [noderef] <asn:3>*<noident>
kernel/sched/psi.c:220:48: sparse: incorrect type in initializer (different address spaces) @@ expected void const [noderef] <asn:3>*__vpp_verify @@ got nst [noderef] <asn:3>*__vpp_verify @@
kernel/sched/psi.c:220:48: expected void const [noderef] <asn:3>*__vpp_verify
kernel/sched/psi.c:220:48: got struct psi_group_cpu *<noident>
kernel/sched/psi.c:256:17: sparse: expression using sizeof(void)
kernel/sched/psi.c:257:17: sparse: expression using sizeof(void)
kernel/sched/psi.c:339:18: sparse: incorrect type in initializer (different address spaces) @@ expected void const [noderef] <asn:3>*__vpp_verify @@ got nst [noderef] <asn:3>*__vpp_verify @@
kernel/sched/psi.c:339:18: expected void const [noderef] <asn:3>*__vpp_verify
kernel/sched/psi.c:339:18: got struct psi_group_cpu *<noident>
>> kernel/sched/psi.c:504:26: sparse: incorrect type in assignment (different address spaces) @@ expected struct psi_group_cpu *cpus @@ got struct psi_group_struct psi_group_cpu *cpus @@
kernel/sched/psi.c:504:26: expected struct psi_group_cpu *cpus
kernel/sched/psi.c:504:26: got struct psi_group_cpu [noderef] <asn:3>*<noident>
>> kernel/sched/psi.c:514:32: sparse: incorrect type in argument 1 (different address spaces) @@ expected void [noderef] <asn:3>*__pdata @@ got oid [noderef] <asn:3>*__pdata @@
kernel/sched/psi.c:514:32: expected void [noderef] <asn:3>*__pdata
kernel/sched/psi.c:514:32: got struct psi_group_cpu *cpus
>> kernel/sched/psi.c:425:21: sparse: dereference of noderef expression
vim +504 kernel/sched/psi.c
330
331 static void psi_group_change(struct psi_group *group, int cpu, u64 now,
332 unsigned int clear, unsigned int set)
333 {
334 enum psi_state state = PSI_NONE;
335 struct psi_group_cpu *groupc;
336 unsigned int *tasks;
337 unsigned int to, bo;
338
> 339 groupc = per_cpu_ptr(group->cpus, cpu);
340 tasks = groupc->tasks;
341
342 /* Update task counts according to the set/clear bitmasks */
343 for (to = 0; (bo = ffs(clear)); to += bo, clear >>= bo) {
344 int idx = to + (bo - 1);
345
346 if (tasks[idx] == 0 && !psi_bug) {
347 printk_deferred(KERN_ERR "psi: task underflow! cpu=%d idx=%d tasks=[%u %u %u] clear=%x set=%x\n",
348 cpu, idx, tasks[0], tasks[1], tasks[2],
349 clear, set);
350 psi_bug = 1;
351 }
352 tasks[idx]--;
353 }
354 for (to = 0; (bo = ffs(set)); to += bo, set >>= bo)
355 tasks[to + (bo - 1)]++;
356
357 /* Time in which tasks wait for the CPU */
358 state = PSI_NONE;
359 if (tasks[NR_RUNNING] > 1)
360 state = PSI_SOME;
361 time_state(&groupc->res[PSI_CPU], state, now);
362
363 /* Time in which tasks wait for memory */
364 state = PSI_NONE;
365 if (tasks[NR_MEMSTALL]) {
366 if (!tasks[NR_RUNNING] ||
367 (cpu_curr(cpu)->flags & PF_MEMSTALL))
368 state = PSI_FULL;
369 else
370 state = PSI_SOME;
371 }
372 time_state(&groupc->res[PSI_MEM], state, now);
373
374 /* Time in which tasks wait for IO */
375 state = PSI_NONE;
376 if (tasks[NR_IOWAIT]) {
377 if (!tasks[NR_RUNNING])
378 state = PSI_FULL;
379 else
380 state = PSI_SOME;
381 }
382 time_state(&groupc->res[PSI_IO], state, now);
383
384 /* Time in which tasks are non-idle, to weigh the CPU in summaries */
385 if (groupc->nonidle)
386 groupc->nonidle_time += now - groupc->nonidle_start;
387 groupc->nonidle = tasks[NR_RUNNING] ||
388 tasks[NR_IOWAIT] || tasks[NR_MEMSTALL];
389 if (groupc->nonidle)
390 groupc->nonidle_start = now;
391
392 /* Kick the stats aggregation worker if it's gone to sleep */
393 if (!delayed_work_pending(&group->clock_work))
394 schedule_delayed_work(&group->clock_work, PSI_FREQ);
395 }
396
397 void psi_task_change(struct task_struct *task, u64 now, int clear, int set)
398 {
399 #ifdef CONFIG_CGROUPS
400 struct cgroup *cgroup, *parent;
401 #endif
402 int cpu = task_cpu(task);
403
404 if (psi_disabled)
405 return;
406
407 if (!task->pid)
408 return;
409
410 if (((task->psi_flags & set) ||
411 (task->psi_flags & clear) != clear) &&
412 !psi_bug) {
413 printk_deferred(KERN_ERR "psi: inconsistent task state! task=%d:%s cpu=%d psi_flags=%x clear=%x set=%x\n",
414 task->pid, task->comm, cpu,
415 task->psi_flags, clear, set);
416 psi_bug = 1;
417 }
418
419 task->psi_flags &= ~clear;
420 task->psi_flags |= set;
421
422 psi_group_change(&psi_system, cpu, now, clear, set);
423
424 #ifdef CONFIG_CGROUPS
> 425 cgroup = task->cgroups->dfl_cgrp;
426 while (cgroup && (parent = cgroup_parent(cgroup))) {
427 struct psi_group *group;
428
429 group = cgroup_psi(cgroup);
430 psi_group_change(group, cpu, now, clear, set);
431
432 cgroup = parent;
433 }
434 #endif
435 }
436
437 /**
438 * psi_memstall_enter - mark the beginning of a memory stall section
439 * @flags: flags to handle nested sections
440 *
441 * Marks the calling task as being stalled due to a lack of memory,
442 * such as waiting for a refault or performing reclaim.
443 */
444 void psi_memstall_enter(unsigned long *flags)
445 {
446 struct rq_flags rf;
447 struct rq *rq;
448
449 if (psi_disabled)
450 return;
451
452 *flags = current->flags & PF_MEMSTALL;
453 if (*flags)
454 return;
455 /*
456 * PF_MEMSTALL setting & accounting needs to be atomic wrt
457 * changes to the task's scheduling state, otherwise we can
458 * race with CPU migration.
459 */
460 rq = this_rq_lock_irq(&rf);
461
462 update_rq_clock(rq);
463
464 current->flags |= PF_MEMSTALL;
465 psi_task_change(current, rq_clock(rq), 0, TSK_MEMSTALL);
466
467 rq_unlock_irq(rq, &rf);
468 }
469
470 /**
471 * psi_memstall_leave - mark the end of an memory stall section
472 * @flags: flags to handle nested memdelay sections
473 *
474 * Marks the calling task as no longer stalled due to lack of memory.
475 */
476 void psi_memstall_leave(unsigned long *flags)
477 {
478 struct rq_flags rf;
479 struct rq *rq;
480
481 if (psi_disabled)
482 return;
483
484 if (*flags)
485 return;
486 /*
487 * PF_MEMSTALL clearing & accounting needs to be atomic wrt
488 * changes to the task's scheduling state, otherwise we could
489 * race with CPU migration.
490 */
491 rq = this_rq_lock_irq(&rf);
492
493 update_rq_clock(rq);
494
495 current->flags &= ~PF_MEMSTALL;
496 psi_task_change(current, rq_clock(rq), TSK_MEMSTALL, 0);
497
498 rq_unlock_irq(rq, &rf);
499 }
500
501 #ifdef CONFIG_CGROUPS
502 int psi_cgroup_alloc(struct cgroup *cgroup)
503 {
> 504 cgroup->psi.cpus = alloc_percpu(struct psi_group_cpu);
505 if (!cgroup->psi.cpus)
506 return -ENOMEM;
507 psi_group_init(&cgroup->psi);
508 return 0;
509 }
510
511 void psi_cgroup_free(struct cgroup *cgroup)
512 {
513 cancel_delayed_work_sync(&cgroup->psi.clock_work);
> 514 free_percpu(cgroup->psi.cpus);
515 }
516
---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation
